Database Reference
In-Depth Information
Additionally, several other works on map-matching have been intro-
duced that assume spatially and temporally high resolution GPS data
is available [6, 7, 25, 34]. These models are similar in that they assume
a small degree of error in the observations which allows them to use
relatively simple nearest-neighbor approaches to map the object's GPS
observation to a road segment on the known road network.
3.4 Tracking for External Sensing
The work mentioned up until this point has all assumed that the
mobile objects were providing their location willingly in order to navigate
or be queried. However, once this assumption is removed, the problem
becomes significantly more challenging. At the core of these challenges is
the fact that we do not know which observations belong to which mobile
objects, referred to as the
data association
problem. For instance, if
two objects are nearby, their observations may get switched, thus the
trajectory we obtain would actually be composed of the movements of
two different objects.
Although the data association problem makes external sensing a much
more complicated problem, it is not the only issue in this scenario. Be-
cause sensing occurs in an incredibly noisy environment, we may detect
false positives as well as miss the detection of actual objects (false neg-
atives). Moreover, the total number of mobile objects is considered to
be unknown. New tracking algorithms have been developed for this sce-
nario, making use of particle filtering methods and finite set statistics
(FSS) [42, 53, 61, 64, 86, 85]. The problem scenario of external sens-
ing has not been addressed in the database community, mainly because
the current solutions only scale to managing 5
−
10 objects, as high
dimensional filtering is known to be an open problem.
4. Mining Mobility Data
Querying spatiotemporal data is able to provide answers to simple
questions, such as
what are the closest coffee shops to me?
or
how
many objects have passed through this area over a given time interval
?
However, extracting semantically higher-order information from such low
level data is a dicult problem. For instance, it may be of interest
to identify those mobile objects that behave similarly (e.g. travel to
similar locations), identify popular or ecient routes, or just to be able
to quantitatively characterize and predict user movements.
We partition this section into three major areas that cover recent work
in mining spatiotemporal data: (i) clustering, (ii) route detection, and
(iii) movement patterns. The first, covers work on clustering moving
